Google's $12.2 billion 'stake' in Marvell is a warrant it hasn't paid for. The vesting schedule is the real contract.

The number that moved on Tuesday was $12.2 billion, and like most numbers that move a stock eight percent before lunch, it was doing more advertising than accounting. That is the headline value of the stake Alphabet's Google can take in Marvell Technology — up to 58.97 million shares at $206.58 apiece — under a custom-silicon agreement the two companies unveiled on August 19. Marvell rose. Broadcom, the incumbent in the same seat, fell. The word attached to the story in every write-up was 'stake.' It is the wrong word, and the right word changes what the deal is.

What Google actually received is a warrant: the right, not the obligation, to buy those shares at a fixed price. It has paid nothing for the stock and owns none of it today. More to the point, most of that right does not yet exist. The warrant vests — becomes exercisable — only as Google meets agreed purchasing targets from Marvell through fiscal 2033. The equity is not a payment for anything. It is a reward schedule, and the thing it rewards is Google's own spending.

The mechanism: an option you earn by buying

Start with the instrument, because the instrument is the argument. A warrant is a call option a company writes on its own shares and hands to someone it wants something from. This one carries a strike of $206.58. If Marvell trades above that level, the warrant is worth exercising and Google pockets the difference between the market price and the strike, multiplied across nearly 59 million shares. If Marvell trades below it, the warrant is a piece of paper. So far this is ordinary corporate plumbing. The unusual part is the condition bolted to the front of it.

Most of the warrant does not vest on a calendar. It vests on a purchase order. Google unlocks the right to those shares in tranches as it hits spending milestones with Marvell, running out to fiscal 2033. Read that the way a treasurer would, not a headline writer. Marvell has handed its largest prospective customer a financial incentive to concentrate its orders in Marvell's direction and keep them there for the better part of a decade. The more Google buys, the more of the warrant vests; the more it vests, the larger the paper gain Google is sitting on; the larger the gain, the stronger the reason to buy again. It is a flywheel assembled out of a customer's own procurement, and it is built to spin one way.

That is why the $12.2 billion is the least interesting number in the release. The figure the warrant exists to defend is the other one Marvell has floated: roughly $120 billion of potential revenue through fiscal 2033, if Google hits the targets the stock option depends on. The $12.2 billion is the carrot. The $120 billion is the field it is standing in. One is the incentive; the other is the outcome the incentive is engineered to produce, and the whole structure only makes sense when you read them together.

The stake is not what Google is paying. It is what Google is being paid to keep buying. — The trade

What the money is actually for

The agreement is broad in a way that tells you it is strategic rather than transactional. Marvell will develop a suite of technologies for Google's TPU ecosystem: AI inference accelerators, data-storage controllers, networking silicon, memory-interface controllers, and near-memory computing. That is not one chip. That is a standing engineering relationship around the accelerators Google uses to train and, increasingly, to serve its models — the in-house alternative that keeps Google from renting all of its compute from Nvidia at Nvidia's margin. Custom silicon is Google's lever against the one supplier every hyperscaler is trying to depend on a little less each quarter.

Broadcom has been the marquee name in that lever for years, the partner most associated with Google's TPU designs. So the reflex reading of Tuesday was displacement, and the tape agreed for a few hours: Marvell up sharply, Broadcom down three to five percent on a deal it was not part of. Morningstar's William Kerwin pushed back on the reflex, calling it 'a big win for Marvell' but framing it as 'a growing pie at Google for new sources, rather than a competitive displacement of Broadcom.' He is probably right on the mechanics. It is also beside the point. Whether this is a bigger pie or a smaller slice for the incumbent, the tell is the same: Google felt it worth handing a second supplier equity to make sure the pie keeps growing on Google's terms.

Vendor financing, run in reverse

To see what is new here, hold it next to the deals it rhymes with. The dominant financial pattern of this AI cycle has been the supplier financing its own demand: Nvidia lining up a reported half a trillion dollars of third-party capital to help customers buy Nvidia systems, chipmakers taking warrants in the labs that buy their parts, capacity underwritten by the vendor that sells into it. The money flows from the seller toward the buyer so the buyer can afford to become a bigger buyer. It keeps the order book full and the story intact.

Google–Marvell flips the arrows. Here the customer takes the equity in the supplier, and ties the value of that equity to how much the customer spends. Different direction, identical anxiety. In both structures, demand and supply in AI are no longer simply discovered in a market where a buyer needs a part and pays for it. They are manufactured, in advance and on paper, by aligning two balance sheets so tightly that neither can easily walk away. When companies start using their share registers to guarantee their order books, the interesting question is no longer how good the chips are. It is what the arrangement is compensating for.

The concentration nobody put on the slide

Concentration is the risk that compounds quietly and arrives all at once, and this deal manufactures it on both sides of the same name. Marvell has just repriced its growth story around one hyperscaler's multi-year purchasing. If Google fully exercises, it becomes Marvell's fifth-largest shareholder. It is already positioned to be one of Marvell's largest custom-silicon customers. That means Marvell is now long Google in its order book and its cap table at the same time — two exposures to a single counterparty, pointing the same way, dressed up as a partnership. It is not a partnership. It is a position, and positions have a downside the announcement did not mention.

Google, by contrast, has bought cheap optionality and paid for it with a promise. It committed no cash for the stock, secured a second source of custom silicon away from Broadcom, gained a lever over a key supplier's roadmap, and added one more hedge against Nvidia — all for a warrant it only funds by spending money it was likely to spend anyway. If the AI capex boom that underwrites the whole thing rolls over, the warrant simply fails to vest and quietly expires. Google is out nothing. Marvell is out the growth multiple the market handed it on Tuesday. The party that structured the deal is the party that priced its own downside out of it.

Marvell is long Google in its order book and its share register at once. That is not a partnership. It is a position — and positions have a downside the press release skips.

What's priced in, and what isn't

The market did what markets do with a clean story: it priced the upside promptly and the fine print not at all. An eight percent move says investors read 'growing pie at Google' and bought it. What that move does not contain is the contingency threaded through every clause. The revenue is a target, not a contract. The equity is a warrant, not a holding. The relationship's value rests on purchase milestones stretching to fiscal 2033 — an eternity in a business where a model architecture can shift the demand curve in a single release. If those targets slip, because Google's own capex cools, or Google designs more in-house, or a renegotiation trims the price, the warrant under-vests and the $120 billion shrinks toward a number no one is modelling. And none of that surfaces on a chart until the quarter it does.

So take the deal for what it is rather than what it announced. It is a genuine win for Marvell, a smart and nearly costless move by Google, and a small landmark in a market that has started using equity to buy its own demand. It is also a study in who holds the leash. Google gave up nothing it cannot recover and took a claim on a supplier's growth; Marvell gave up a measure of its independence to earn a number it does not yet possess. The best case here is spectacular and entirely plausible. But the trade and the company are, as usual, two different things, and the distance between them — a warrant that only pays if the buyer keeps buying — is the part worth watching after everyone else has moved on.